Normal Extensions of Linear Relations via Quaternionic Cayley Transforms
F.-H. Vasilescu
Department of Mathematics University of Lille 1
June 29, 2013, Alba Iulia
Outline
1 Introduction
Notation and Preliminaries Normal Relations
2 Quaternionic Cayley Transforms of Linear Relations E–Cayley Transform
InverseE–Cayley Transform
3 Unitary Operators and the Inverse Quaternionic Cayley Transform
Special Unitary Operators
Other special Classes of Operator and Relations
4 Normal Extensions Subnormality
Abstract
In a previous work, the author has introduced a notion of quaternionic Cayley transform, valid for some classes of pairs of symmetric operators. In a second work, the former definition was slightly modified, leading to more direct proofs, using von Neumann’s Cayley transform.
In the present work, written in cooperation withAdrian Sandovici, a quaternionic Cayley transform for linear relations is introduced and some of its properties are exhibited. We emphasize the role played by the linear relations whose quaternionic Cayley transforms are unitary operators, which happen to be normal relations, and investigate the class of those linear relations which extend
Abstract
In a previous work, the author has introduced a notion of quaternionic Cayley transform, valid for some classes of pairs of symmetric operators. In a second work, the former definition was slightly modified, leading to more direct proofs, using von Neumann’s Cayley transform.
In the present work, written in cooperation withAdrian Sandovici, a quaternionic Cayley transform for linear relations is introduced and some of its properties are exhibited. We emphasize the role played by the linear relations whose quaternionic Cayley transforms are unitary operators, which happen to be normal relations, and investigate the class of those linear relations which extend
Some References
R. Arens,Operational calculus of linear relations, Pacific J.
Math. 11 (1961) 9-23.
E. A. Coddington,Extension theory of formally normal and symmetric subspaces, Mem. Amer. Math. Soc., 134, 1973.
A. Sandovici and F.-H. Vasilescu,Normal Extensions of Subnormal Linear Relations via Quaternionic Cayley Transforms, Monatsh. Math. (to appear)
F.-H. Vasilescu,Quaternionic Cayley transform, J.
Functional Analysis164(1999), 134-162.
F.-H. Vasilescu,Quaternionic Cayley transform revisited,
Introduction
Let(H,h·,·i)be a complex Hilbert space and letU be a unitary operator acting inH. IfIis the identity operator onH, set
S={{(I−U)x,i(I+U)x};x ∈ H}, (1) which is a linear subspace ofH2=H × H. The spaceSis the graph of a linear transformation inHif and only if the operator I−U is injective. In this case, the corresponding linear
transformation is known as theinverse Cayley transformof the unitary operatorU, which is, in general, a (not necessarily bounded) self-adjoint operator.
Nevertheless, the spaceSgiven by (1) is well defined without the conditionI−U invertible, and it is what is called as alinear relationinH(followingArens). Moreover, we have
S∗:={{u,v} ∈ H2:hx,vi=hy,ui, ∀ {x,y} ∈S}=S, whereS∗ stands for the adjoint relation ofS. In other words,S is aself-adjointlinear relation (formal definitions will be later given).
We shall define a Cayley transform for some linear relations.
Briefly, considering a linear relationSinH2(that is, a linear subspace ofH2× H2), we associate it with the linear relation
V ={{{x10,x20}+{ix1,−ix2},{x10,x20}+{−ix1,ix2}} : {{x1,x10},{x2,x20}} ∈S}, which is a quaternionic type Cayley transform ofS. Under some natural conditions, the linear spaceV is the graph of a partial isometry inH2. Our aim is to study various properties relating the linear relationSand the partial isometryV.
An "extreme" situation is whenU is an operator of the form U =
T iA iA T∗
,
withT normal andAself-adjoint inH, such thatTT∗+A2=I andAT =TA.In this case,U is unitary, and it is a quaternionic Cayley transform of the linear relation
{{i(T−I)x−Ay,Ax−i(T∗−I)y},(T+I)x+iAy,iAx+ (T∗+I)y}, {x,y} ∈ H2},
which is normal (in a sense to be defined).
Notation and Preliminaries
LetHbe a complex Hilbert space. Alinear relation(briefly, alr) inHis a vector subspace ofH2=H × H. The elements ofH2 are represented as pairs{x,y}, withx,y ∈ H.
For alr T inH, we denote byD(T),R(T),N(T),M(T)its domain of definition, itsrange, itskernel, and itsmultivalued part, respectively.
We often identify a linear mapS:D(S)⊂ H 7→ Hwith its graph G(S).
Notation and Preliminaries
LetHbe a complex Hilbert space. Alinear relation(briefly, alr) inHis a vector subspace ofH2=H × H. The elements ofH2 are represented as pairs{x,y}, withx,y ∈ H.
For alr T inH, we denote byD(T),R(T),N(T),M(T)its domain of definition, itsrange, itskernel, and itsmultivalued part, respectively.
We often identify a linear mapS:D(S)⊂ H 7→ Hwith its graph G(S).
IfS,T arelr inH2, theircompositionis given by ST ={{x,z};∃y :{x,y} ∈T,{y,z} ∈S}.
Theadjoint T∗ ofT is given by
T∗ ={{u,v}:hu,yi=hv,xi;∀{x,y} ∈T}.
IfT ⊂T∗,T is said to besymmetric.
IfT =T∗,T is said to beself-adjoint.
IfS,T arelr inH2, theircompositionis given by ST ={{x,z};∃y :{x,y} ∈T,{y,z} ∈S}.
Theadjoint T∗ ofT is given by
T∗ ={{u,v}:hu,yi=hv,xi;∀{x,y} ∈T}.
IfT ⊂T∗,T is said to besymmetric.
IfT =T∗,T is said to beself-adjoint.
IfS,T arelr inH2, theircompositionis given by ST ={{x,z};∃y :{x,y} ∈T,{y,z} ∈S}.
Theadjoint T∗ ofT is given by
T∗ ={{u,v}:hu,yi=hv,xi;∀{x,y} ∈T}.
IfT ⊂T∗,T is said to besymmetric.
IfT =T∗,T is said to beself-adjoint.
IfS,T arelr inH2, theircompositionis given by ST ={{x,z};∃y :{x,y} ∈T,{y,z} ∈S}.
Theadjoint T∗ ofT is given by
T∗ ={{u,v}:hu,yi=hv,xi;∀{x,y} ∈T}.
IfT ⊂T∗,T is said to besymmetric.
IfT =T∗,T is said to beself-adjoint.
Normal Relations
A linear relationSin a Hilbert spaceHis said to beformally normal(Coddington) if there exists an isometryV :S →S∗ of the form
V{x,x0}={x,x00}, {x,x0} ∈S, {x,x00} ∈S∗. (2) In particular,kx0k=kx00k.
A formally normal linear relationS inHis said to benormal (Coddington) if the isometryV is surjective.
Normal relations are automatically closed.
Normal Relations
A linear relationSin a Hilbert spaceHis said to beformally normal(Coddington) if there exists an isometryV :S →S∗ of the form
V{x,x0}={x,x00}, {x,x0} ∈S, {x,x00} ∈S∗. (2) In particular,kx0k=kx00k.
A formally normal linear relationS inHis said to benormal (Coddington) if the isometryV is surjective.
Normal relations are automatically closed.
Normal Relations
A linear relationSin a Hilbert spaceHis said to beformally normal(Coddington) if there exists an isometryV :S →S∗ of the form
V{x,x0}={x,x00}, {x,x0} ∈S, {x,x00} ∈S∗. (2) In particular,kx0k=kx00k.
A formally normal linear relationS inHis said to benormal (Coddington) if the isometryV is surjective.
Normal relations are automatically closed.
Characterization of Normality
As in the case of (unbounded) normal operators, we have the following:
Theorem (Sandovici+V)
LetSbe a linear relation in a Hilbert spaceH. Equivalent are:
1 Sis a normal linear relation inH;
2 Sis closed,D(S) =D(S∗), andS∗S=SS∗.
Quaternionic Cayley Transform of Linear Relations
LetSbe a symmetric linear relation inH. We define the relation V ={{x0+ix,x0−ix};{x,x0} ∈S},
which is called theCayley transformofS(Arens).
BecauseSis symmetric, we have
kx0±ixk2=kxk2+kx0k2, {x,x0} ∈S.
Therefore,V is (the graph of) an isometry, defined on D(V) ={x0+ix,;{x,x0} ∈S} ⊂ H, whose range is R(V) ={x0−ix;{x,x0} ∈S} ⊂ H.
Quaternionic Cayley Transform of Linear Relations
LetSbe a symmetric linear relation inH. We define the relation V ={{x0+ix,x0−ix};{x,x0} ∈S},
which is called theCayley transformofS(Arens).
BecauseSis symmetric, we have
kx0±ixk2=kxk2+kx0k2, {x,x0} ∈S.
Therefore,V is (the graph of) an isometry, defined on D(V) ={x0+ix,;{x,x0} ∈S} ⊂ H, whose range is R(V) ={x0−ix;{x,x0} ∈S} ⊂ H.
Version of a von Neumann’s Theorem
The next result was proved byArens:
TheoremLetS be a symmetric relation in the Hilbert spaceH.
The relationS is self-adjoint if and only if the Cayley transform ofSis a unitary operator inH.
Recall: Algebra of Quaternions
Consider the 2×2-matrices I=
1 0 0 1
, J=
1 0 0 −1
,
K=
0 1
−1 0
, L=
0 1 1 0
.
The Hamilton algebra of quaternionsHcan be identified with theR-subalgebra of the algebraM2of 2×2-matrices with complex entries, generated by the matricesI,iJ,KandiL. This allows us to regard the elements ofHas matrices and to perform some operations inM2rather than inH.
Recall: Some Elementary Operations
We haveJ∗=J,K∗ =−K,L∗=L,J2=−K2=L2=I, JK=L=−KJ,KL=J=−LK,JL=K=−LJ,where the adjoints are computed in the Hilbert spaceC2.
We also putE=iJ,F=iL, and we we haveE∗ =−E,E2=−I, F∗=−F,F2=−I
A Preliminary Lemma
LetHbe a complex Hilbert space. The matrices fromM2
clearly act onH2(whose natural norm is denoted byk ∗ k2).
Lemma
LetSbe a linear relation inH2. Suppose that the linear relation JSis symmetric. Then we have
kx0±Exk22=kx0k22+kxk22, {x,x0} ∈S. (3)
E–Cayley Transform
LetSbe a linear relation inH2such thatJS is symmetric.
E–Cayley transform ofS:
V :=
{x0+Ex,x0−Ex} : {x,x0} ∈S . (4) As we havekx0+Exk2=kx0−Exk2, it follows thatV is (the graph of) a partial isometry withD(V) =R(S+E)and R(V) =R(S−E).
In fact,V given by
V :R(S+E)7→R(S−E), V(x0+Ex) =x0−Ex, {x,x0} ∈S.
E–Cayley Transform
LetSbe a linear relation inH2such thatJS is symmetric.
E–Cayley transform ofS:
V :=
{x0+Ex,x0−Ex} : {x,x0} ∈S . (4) As we havekx0+Exk2=kx0−Exk2, it follows thatV is (the graph of) a partial isometry withD(V) =R(S+E)and R(V) =R(S−E).
In fact,V given by
V :R(S+E)7→R(S−E), V(x0+Ex) =x0−Ex, {x,x0} ∈S.
E–Cayley Transform
LetSbe a linear relation inH2such thatJS is symmetric.
E–Cayley transform ofS:
V :=
{x0+Ex,x0−Ex} : {x,x0} ∈S . (4) As we havekx0+Exk2=kx0−Exk2, it follows thatV is (the graph of) a partial isometry withD(V) =R(S+E)and R(V) =R(S−E).
In fact,V given by
V :R(S+E)7→R(S−E), V(x0+Ex) =x0−Ex, {x,x0} ∈S.
F–Cayley Transform
IfLSis symmetric, a similar discussion leads to the definition of an operatorW given by
W :R(S+F)7→R(S−F), W(x0+Fx) =x0−Fx, {x,x0} ∈S, which is again a partial isometry. The operatorW is called the F–Cayley transformofS.
Because the two Cayley transforms defined above are alike, in the sequel we shall mainly deal with theE–Cayley transform.
Properties of the E–Cayley Transform
Lemma
LetSbe a linear relation inH2such thatJS is symmetric, and letV be theE–Cayley transform ofS. We have the following:
(a)V is closed if and only ifSis closed, and if and only if the spacesR(S±E)are closed;
(b)One hasN(I−V) =M(S). The operatorI−V is injective if and only ifSis an operator. Moreover,Sis densely defined if and only if the spaceR(I−V)is dense inH2;
(c)IfSK⊂KS, thenSK=KSandV−1=−KVK;
(d)The lrJSis self-adjoint if and only ifV is unitary inH2. (e)TheE–Cayley transform is an inclusion preserving map.
Inverse E–Cayley Transform
LetV :D(V)⊂ H27→ H2be a partial isometry, for which we define the following:
Inverse E–Cayley transform:
S :={{E(V −I)x,(V +I)x} : x ∈D(V)}, which is a linear relation inH2.
In a similar way, we can define theinverseF–Cayley transform.
These two (quaternionic) inverse Cayley transforms have similar properties, and so we shall deal only with the inverse E–Cayley transform.
Inverse E–Cayley Transform
LetV :D(V)⊂ H27→ H2be a partial isometry, for which we define the following:
Inverse E–Cayley transform:
S :={{E(V −I)x,(V +I)x} : x ∈D(V)}, which is a linear relation inH2.
In a similar way, we can define theinverseF–Cayley transform.
These two (quaternionic) inverse Cayley transforms have similar properties, and so we shall deal only with the inverse E–Cayley transform.
Properties of the Inverse E–Cayley Transform
LemmaLetV :D(V)⊂ H27→ H2be a partial isometry. Then the linear relation
S={{E(V−I)x,(V +I)x} : x ∈D(V)}. has the following properties:
(i)the linear relationJSis symmetric and theE–Cayley transform ofSisV;
(ii)we haveV−1=−KVKif and only ifSK=KS.
Global Properties
We summarize the properties of the quaternionic Cayley transform:
Theorem
TheE-Cayley transform is an inclusion preserving bijective map assigning to each linear relationSinH2withJS symmetric a partial isometryV inH2. Moreover:
(1)the operatorV is closed if and only if the linear relationSis closed;
(2)the equalityV−1=−KVKholds if and only if the equality SK=KSholds;
(3)the linear relationJSis self-adjoint if and only ifV is unitary onH2.
Special Unitary Operators
In this this section we are particularly interested in those unitary operators producing normal linear relations (in particular
(unbounded) normal operators), via the inverseE–Cayley transform.
LemmaLetUbe a bounded operator onH2. The operatorUis unitary and has the propertyU∗ =−KUKif and only if there are a bounded operatorT and bounded self-adjoint operators A,B onHsuch thatTT∗+A2=I,T∗T +B2=I,AT =TBand
U=
T iA iB T∗
, whereIthe identity onH.
Some Auxilliary Results
LemmaLetUbe a unitary operator onH2withU∗ =−KUK. If we set
S={{E(U−I)x,(U+I)x} : x ∈ H}. we have thatSis a closed linear relation and
S∗ ={{(I−U)x,E(I+U)x} : x ∈ H}.
LemmaLetUbe an operator onH2having the form U=
T iA iB T∗
,
withT,A=A∗,B=B∗ bounded operators onH, such that
∗ 2 ∗ 2
Intertwining Isometries
LemmaLetV be a partial isometry such thatV−1=−KVK.
LetSbe the inverseE–Cayley transform ofV. Equivalent are:
(i)JD(S)⊂D(S)and there exists an isometryH :S→Sof the formH{x,x0}={Jx,x00};
(ii)there exists an isometryG:D(V)7→D(V)such that E(I−V) = (I−V)G.
Remark
(1) The isometryGgiven by the previous Lemma is not uniquely determined. IfG1,G2are two isometries as in this lemma, then the operatorG1−G2is clearly defined onD(V) and has values in the kernel ofI−V.
(2) Assume that the isometryHin the previous Lemma is surjective. Then it can be shown that the isometric operatorG may be chosen to be surjective. Conversely, ifGis surjective, the operatorH may be also chosen to be surjective.
Proposition
LetUbe a unitary operator onH2with the property
U∗=−KUK. Let alsoSbe the inverseE–Cayley transform of U. The linear relationSis normal if and only is there exists a unitary operatorGU onH2such thatE(I−U) = (I−U)GU and (GU)∗ =−GU.
Main result and the Class U
C(H
2)
Theorem
LetUbe a unitary operator onH2with the property
U∗=−KUK, and letS be the inverseE–Cayley transform ofU.
The linear relationSis normal if and only if (U+U∗)E=E(U+U∗).
DefinitionLetU(H2)be the set of all unitary operators inH2. We also set
UC(H2) ={U ∈ U(H2);U∗=−KUK, (U+U∗)E=E(U+U∗)},
Main result and the Class U
C(H
2)
Theorem
LetUbe a unitary operator onH2with the property
U∗=−KUK, and letS be the inverseE–Cayley transform ofU.
The linear relationSis normal if and only if (U+U∗)E=E(U+U∗).
DefinitionLetU(H2)be the set of all unitary operators inH2. We also set
UC(H2) ={U ∈ U(H2);U∗=−KUK, (U+U∗)E=E(U+U∗)},
The Class N
IC(H
2)
DefinitionWe introduce the following class of linear relations in H2:
NIC(H2) ={S ⊂ H2× H2; S normal,(JS)∗ =JS,KS=SK}.
RemarkThe previous results show that the map
NIC(H2)3S7→US ∈ UC(H2),
whereUS stands for theE–Cayley transform ofS, is bijective.
The Class N
IC(H
2)
DefinitionWe introduce the following class of linear relations in H2:
NIC(H2) ={S ⊂ H2× H2; S normal,(JS)∗ =JS,KS=SK}.
RemarkThe previous results show that the map NIC(H2)3S7→US ∈ UC(H2),
whereUS stands for theE–Cayley transform ofS, is bijective.
Subnormality
Adapting the terminology from operator theory, we may say that a linear relationSin a Hilbert spaceHissubnormalif there exists a Hilbert spaceK ⊃ Hand a normal relationN inK such thatS ⊂N. Nevertheless, we are particularly interested to solve such a problem with the restrictionK=H. In fact, to apply the quaternionic Cayley transform machinery, we consider only linear relations in the Hilbert spaceH2.
The Class S
IC(H
2)
We start by introducing a class of linear relations having necessary properties connected to subnormality.
LetT :be a linear relation inH2, withD(T) =D0⊕D0,D0⊂ H, which is equivalent to the inclusions
(i)JD(T)⊂D(T)andKD(T)⊂D(T).
Furthermore, assume thatT satisfies the following conditions:
(ii)JT is symmetric;
(iii)TK=KT;
(iv)there exists a surjective isometryHT :T →T of the form HT{x,x0}={Jx,x00}; for all{x,x0} ∈T.
Denote bySIC(H2)the set of those linear relationsT inH2 such that(i)–(iv)hold.
The Class P
C(H
2)
LetPC(H2)be the set of those partial isometries V :D(V)⊂ H27→ H2such that:
(a)V−1=−KVK;
(b)ER(I−V) =R(I−V);
(c)there exists a surjective isometryG:D(V)7→D(V)such thatE(I−V) = (I−V)G.
It was proved that theE–Cayley transform is a bijective map fromSIC(H2)ontoPC(H2). Note also thatUC(H2)⊂ PC(H2)by some of the previous results.
Normal Extensions
A decomposition result:
PropositionLetU ∈ UC(H2)and letD ⊂ H2be a closed subspace with the propertiesKU(D)⊂ Dand
E(I−U)(D)⊂(I−U)(D). IfV =U|D,E =D⊥andW =U|E, thenU=V⊕W andV,W ∈ PC(H2).
Main extension result:
Theorem
LetT ∈ SIC(H2). Equivalent are:
(i)the linear relationT has an extensionS inNIC(H2)such thatHT =HS|T;
(ii)there exists aW ∈ PC(H2), withD(W) =R(T +E)⊥.
Normal Extensions
A decomposition result:
PropositionLetU ∈ UC(H2)and letD ⊂ H2be a closed subspace with the propertiesKU(D)⊂ Dand
E(I−U)(D)⊂(I−U)(D). IfV =U|D,E =D⊥andW =U|E, thenU=V⊕W andV,W ∈ PC(H2).
Main extension result:
Theorem
LetT ∈ SIC(H2). Equivalent are:
(i)the linear relationT has an extensionS inNIC(H2)such thatHT =HS|T;
(ii)there exists aW ∈ PC(H2), withD(W) =R(T +E)⊥.
The next assertion provides an extension result for linear relations. In particular, this includes the case of not necessarily densely defined operators.
Corollary
LetT ∈ SIC(H2)be closed and letV be theE–Cayley
transform ofT. Then the linear relationT has an extension in NIC(H2)if and only if there exists aW ∈ PC(H2), with the propertyD(W) =R(T +E)⊥.